PostieDay 37 · 6 October 2026
Chapter 2 · 34 days to the chapter deadline
- Interaction Options for Agents
Following on from yesterday where I went for "Expo over IMAP" I spent today sketching out a plan for a CLI which agents can use to interact with the Postie CLI. Of course this can be used by humans but the CLI is meant for things like automation scripts and agents interactions.
Again, one of the main focuses of Postie is collaboration between Humans, teams and Agents. I dislike the idea of giving agents a browser to interact with Postie (maybe it's because I'm token conscious or whatever) but it just seems less than ideal. As agents are known to get confused and hallucinate I find a good command line interface gives them the structure interact properly.
The idea is that this will include structured JSON outputs and such. It will also split admin functions from general Postie-Mail functions. I can imagine that this being the best way to backup things regularly too. I get really excited at the possibilities for this.
While I spent today sketching out the general features the actual process of this CLI specifically won't continue until I have made a robust permissions system and groups interaction system with API docs as a backup. The last thing I want is AI agents having unstructured, unregulated access to a given Postie account, whether their own or another.
Building from solid, secure foundations is paramount. Don't want to have problems later especially with these indeterministic chaos agents. Yet another thing for me to maintain...